冗余人工肌肉驱动的仿生机器人强化学习控制OA

Artificial muscles are key actuation components for bioinspired robots.However,their current applica-tions remain far from the capabilities of biological muscle systems,particularly due to the lack of redundant and co-ordinated multi-muscle actuation similar to that found in living organisms.To address the challenge of complex ar-tificial-muscle actuation and coordination in bioinspired robots,this study proposes a soft robotic design driven by multiple artificial muscles arranged in parallel and develops a reinforcement learning-based locomotion control strategy for this design.A prototype was developed using a flexible cross-shaped printed circuit board as the main body,integrating six liquid crystal elastomer artificial muscles and their driving circuits.Its strain characteristics and dynamic response performance were experimentally characterized.Considering the deformation and locomotion characteristics of the prototype,a simplified tendon-driven model was constructed in a simulation environment.By properly designing the state space,action space,and reward functions,parallel reinforcement learning training was conducted using the soft Actor-Critic algorithm to obtain coordinated muscle activation strategies for translational and rotational locomotion.The stable periodic segments of the learned locomotion policies were then extracted and used to drive the physical prototype offline.The robot achieved effective multidirectional translation and rotation,demonstrating the feasibility of using reinforcement learning to control complex artificial-muscle-driven systems.